On Wed, Jul 12, 2017 at 11:09:23AM -0700, Jonathan Nieder wrote: > > I didn't follow the rest of the "struct repository" series closely, but > > I don't feel like we ever reached a resolution on how config would be > > handled. I notice that the in-process "ls-files" behaves differently > > than the old one when config differs between the submodule and the > > parent repository. As we convert more commands (that use more config) > > this will become more likely to be noticed by somebody. > > > > Do we have a plan for dealing with this? Is our solution just "recursed > > operations always respect the parent config, deal with it"? > > For settings like branch.<name>.remote, I don't think anyone would > disagree that the right thing to do is to use the per-repository > config of the submodule. The repository object is already able to > handle per-repository config, so this just involves callers being > careful not to cache values locally in a way that conflates > repositories. It should be pretty straightforward (for commands like > "git fetch --recurse-submodules", for example). I agree that's the right approach. What I'm worried about is that I see in-process work proceeding without the "callers being careful" part being audited, which can lead to regressions (e.g., ls-files with core.quotepath is "broken" in next right now). Though at least the regression would be limited to people using submodules. > For settings like grep.patternType, on the other hand, it would be > very strange for the behavior to change when grep crosses the > submodule boundary. So I think using the parent project config is the > right thing to do and the old behavior was simply wrong. In other > words, I don't think this is so much a case of "deal with it" as > "sorry we got the behavior so wrong before --- we've finally fixed it > now". I think that's not actually about the parent project's config, as much as it is about the parameters of the current operation. I.e., the argument is that this particular grep operation is using a particular pattern type, no matter how we arrived at that decision, and it should be used for all of the recursive bits. So whether the superproject has its own grep.patternType set, or whether the user used "-P" on the command line, the result is the same: we need to tell the submodule to ignore any config and use the parameters we feed it. In a multi-process model, that should happen by converting all of the bits in "struct grep_opt" back into command-line parameters and feeding them to the recursive processes (which would then give them precedence over any config). But I'm pretty sure we don't do that. In the in-process model, that would hopefully be a bit simpler, as we'd just pass in a pre-made grep_opt. -Peff
